summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--firmware/drivers/ata.c110
1 files changed, 103 insertions, 7 deletions
diff --git a/firmware/drivers/ata.c b/firmware/drivers/ata.c
index 63e603f76f..b6292e5d5c 100644
--- a/firmware/drivers/ata.c
+++ b/firmware/drivers/ata.c
@@ -162,33 +162,129 @@ static int wait_for_end_of_transfer(void)
162} 162}
163 163
164 164
165/*
1660x090156A8: 0x4F22 sts.l pr,@-r15
1670x090156AA: 0x6243 mov r4,r2
1680x090156AC: 0x6023 mov r2,r0
1690x090156AE: 0xC901 and #0x01,r0
1700x090156B0: 0x2008 tst r0,r0
1710x090156B2: 0x8911 bt 0x090156D8
1720x090156B4: 0x6153 mov r5,r1
1730x090156B6: 0x311C add r1,r1
1740x090156B8: 0x6523 mov r2,r5
1750x090156BA: 0x351C add r1,r5
1760x090156BC: 0xD30E mov.l @(0x03C,pc),r3 ; 0x090156F8 (0x06104100)
1770x090156BE: 0x0009 nop
178
1790x090156C0: 0x6131 mov.w @r3,r1
1800x090156C2: 0x611D extu.w r1,r1
1810x090156C4: 0x2210 mov.b r1,@r2
1820x090156C6: 0x7201 add #0x01,r2
1830x090156C8: 0x4119 shlr8 r1
1840x090156CA: 0x2210 mov.b r1,@r2
1850x090156CC: 0x7201 add #0x01,r2
1860x090156CE: 0x3252 cmp/hs r5,r2
1870x090156D0: 0x8BF6 bf 0x090156C0
188
1890x090156D2: 0xA00F bra 0x090156F4
1900x090156D4: 0x4F26 lds.l @r15+,pr
1910x090156D6: 0x0009 nop
1920x090156D8: 0x6423 mov r2,r4
1930x090156DA: 0x6153 mov r5,r1
1940x090156DC: 0x311C add r1,r1
1950x090156DE: 0x6543 mov r4,r5
1960x090156E0: 0x351C add r1,r5
1970x090156E2: 0xD205 mov.l @(0x018,pc),r2 ; 0x090156F8 (0x06104100)
198
1990x090156E4: 0x6121 mov.w @r2,r1
2000x090156E6: 0x611F exts.w r1,r1
2010x090156E8: 0x6118 swap.b r1,r1
2020x090156EA: 0x2411 mov.w r1,@r4
2030x090156EC: 0x7402 add #0x02,r4
2040x090156EE: 0x3452 cmp/hs r5,r4
2050x090156F0: 0x8BF8 bf 0x090156E4
206
2070x090156F2: 0x4F26 lds.l @r15+,pr
2080x090156F4: 0x000B rts
2090x090156F6: 0x0009 nop
2100x090156F8: 0x0610 .long 0x06104100 ; 0x090156E0
2110x090156FA: 0x4100
212*/
213
214
215
165/* the tight loop of ata_read_sectors(), to avoid the whole in IRAM */ 216/* the tight loop of ata_read_sectors(), to avoid the whole in IRAM */
166static void copy_read_sectors(unsigned char* buf, 217static void copy_read_sectors(unsigned char* buf,
167 int wordcount) 218 int wordcount)
168 __attribute__ ((section (".icode"))); 219 __attribute__ ((section (".icode")));
169static void copy_read_sectors(unsigned char* buf, int wordcount) 220static void copy_read_sectors(unsigned char* buf, int wordcount)
170{ 221{
171 if (wordcount <= 0) 222 unsigned short tmp = 0; /* have to init to prevent warning? */
172 return; /* should never happen, but to protect my tail loop */
173 223
174 if ( (unsigned int)buf & 1 ) 224 if ( (unsigned int)buf & 1)
175 { 225 { /* not 16-bit aligned, copy byte by byte */
176 unsigned char* bufend = buf + wordcount*2; 226 unsigned char* bufend = buf + wordcount*2;
227#ifdef PREFER_C
177 do 228 do
178 { /* loop compiles to 8 assembler instructions */ 229 { /* loop compiles to 9 assembler instructions */
179 unsigned short tmp = ATA_DATA; 230 tmp = ATA_DATA;
180 *buf++ = tmp & 0xff; /* I assume big endian */ 231 *buf++ = tmp & 0xff; /* I assume big endian */
181 *buf++ = tmp >> 8; /* and don't use the SWAB16 macro */ 232 *buf++ = tmp >> 8; /* and don't use the SWAB16 macro */
182 } while (buf < bufend); /* tail loop is faster */ 233 } while (buf < bufend); /* tail loop is faster */
234#else
235 asm (
236 "mov #1, r0 \n"
237 "loop_b: \n"
238 "mov.w @%1,%0 \n"
239 "mov.b %0,@%2 \n"
240 "shlr8 %0 \n"
241 "mov.b %0,@(r0,%2) \n"
242 "add #0x02,%2 \n"
243 "cmp/hs %3,%2 \n"
244 "bf loop_b \n"
245 : /* outputs */
246 : /* inputs */
247 /* %0 */ "r"(tmp),
248 /* %1 */ "r"(&ATA_DATA),
249 /* %2 */ "r"(buf),
250 /* %3 */ "r"(bufend)
251 : /* trashed */
252 "r0"
253 );
254#endif
183 } 255 }
184 else 256 else
185 { 257 { /* 16-bit aligned, can do faster copy */
186 unsigned short* wbuf = (unsigned short*)buf; 258 unsigned short* wbuf = (unsigned short*)buf;
187 unsigned short* wbufend = wbuf + wordcount; 259 unsigned short* wbufend = wbuf + wordcount;
260#ifdef PREFER_C
188 do 261 do
189 { /* loop compiles to 7 assembler instructions */ 262 { /* loop compiles to 7 assembler instructions */
190 *wbuf = SWAB16(ATA_DATA); 263 *wbuf = SWAB16(ATA_DATA);
191 } while (++wbuf < wbufend); /* tail loop is faster */ 264 } while (++wbuf < wbufend); /* tail loop is faster */
265#else
266 asm (
267 "mov #2, r0 \n"
268 "loop_w: \n"
269 "mov.w @%1,%0 \n"
270 "swap.b %0,%0 \n"
271 "mov.w %0,@%2 \n"
272 "mov.w @%1,%0 \n" /* unrolled, do one more */
273 "swap.b %0,%0 \n"
274 "mov.w %0,@(r0,%2) \n"
275 "add #0x04,%2 \n"
276 "cmp/hs %3,%2 \n"
277 "bf loop_w \n"
278 : /* outputs */
279 : /* inputs */
280 /* %0 */ "r"(tmp),
281 /* %1 */ "r"(&ATA_DATA),
282 /* %2 */ "r"(wbuf),
283 /* %3 */ "r"(wbufend)
284 : /* trashed */
285 "r0"
286 );
287#endif
192 } 288 }
193} 289}
194 290